openfilm turns a coding agent into a video agent with an editor
openfilm is an open-source framework that lets any coding agent write a video as a web page and renders it through headless Chromium, with a desktop editor called Studio to change clips, timing, audio, subtitles and layers after the fact.
Image: openfilm, via GitHub
Why it mattersA marketer or founder who already pays for Claude Code or Codex gets a video pipeline that reuses that subscription, so the question moves from which video tool to buy to which model to point at the agent.
A coding agent that writes a web page can be pointed at a video the same way. openfilm, an MIT-licensed project on GitHub, takes any agent that writes code, Claude Code and Codex among them, and gives it a brief, a renderer and an editor so the output is a real video a person can send. The repository is a month old, created on 4 October 2026, and currently carries 440 stars and 52 forks.
What the agent does
A user asks the agent for a video. The agent writes a web page using three.js, WebGL, canvas, SVG, CSS and GSAP, and openfilm renders that page frame by frame through headless Chromium into an MP4, a GIF, a transparent WebM, an audio-only file, a still, or an SRT subtitle track. The video is code, which is the point: anything the agent could change about a webpage, it can change about the video. A missing word in a subtitle is a string edit, a slower clip is a timing variable, a different voice is one call to a different service.
Studio, where a person fixes the result
The project ships with a desktop app called Studio. It opens the generated project and lets a person change text and layers, drag clip timings on a timeline, adjust position, styling, volume and playback speed, rewrite subtitles, and export to the formats above. The install path for a terminal user is npx skills add openfilm/openfilm; a desktop binary is at openfilm.dev. The system needs Node 22.2 or newer, ffmpeg on the path, and downloads around 100 MB of headless Chromium on first run.
Who it calls, and who pays
openfilm itself does not host a model. For voice, image or video generation inside a clip, a user connects their own account for ElevenLabs, OpenAI, Google or fal in Settings, and openfilm reads the keys from there. The same holds for the coding agent: openfilm sits in front of Claude Code or Codex, so the model bill belongs to whichever subscription the user already has. The repository description names the two by name as the supported agents.
A team that already runs a coding agent on a subscription now has a video stack that reuses that subscription and leaves the model choice open. For a small team producing content, that is a different cost shape to a per-minute paid video tool, with the trade that the output is only as good as the model and the prompts behind it. 440 stars in twelve days is real traction for a repository this young, and the design makes the rendering engine, the editor and the model layer three separate pieces, so a project that only wants the subtitle export or the Studio editor can keep that and swap out the parts it does not need.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.